xref: /openbmc/linux/fs/ext4/namei.c (revision 39b6f3aa)
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *	Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *	Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *	Theodore Ts'o, 2002
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39 
40 #include "xattr.h"
41 #include "acl.h"
42 
43 #include <trace/events/ext4.h>
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 
51 static struct buffer_head *ext4_append(handle_t *handle,
52 					struct inode *inode,
53 					ext4_lblk_t *block)
54 {
55 	struct buffer_head *bh;
56 	int err = 0;
57 
58 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 		     ((inode->i_size >> 10) >=
60 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 		return ERR_PTR(-ENOSPC);
62 
63 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 
65 	bh = ext4_bread(handle, inode, *block, 1, &err);
66 	if (!bh)
67 		return ERR_PTR(err);
68 	inode->i_size += inode->i_sb->s_blocksize;
69 	EXT4_I(inode)->i_disksize = inode->i_size;
70 	err = ext4_journal_get_write_access(handle, bh);
71 	if (err) {
72 		brelse(bh);
73 		ext4_std_error(inode->i_sb, err);
74 		return ERR_PTR(err);
75 	}
76 	return bh;
77 }
78 
79 static int ext4_dx_csum_verify(struct inode *inode,
80 			       struct ext4_dir_entry *dirent);
81 
82 typedef enum {
83 	EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85 
86 #define ext4_read_dirblock(inode, block, type) \
87 	__ext4_read_dirblock((inode), (block), (type), __LINE__)
88 
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 					      ext4_lblk_t block,
91 					      dirblock_type_t type,
92 					      unsigned int line)
93 {
94 	struct buffer_head *bh;
95 	struct ext4_dir_entry *dirent;
96 	int err = 0, is_dx_block = 0;
97 
98 	bh = ext4_bread(NULL, inode, block, 0, &err);
99 	if (!bh) {
100 		if (err == 0) {
101 			ext4_error_inode(inode, __func__, line, block,
102 					       "Directory hole found");
103 			return ERR_PTR(-EIO);
104 		}
105 		__ext4_warning(inode->i_sb, __func__, line,
106 			       "error reading directory block "
107 			       "(ino %lu, block %lu)", inode->i_ino,
108 			       (unsigned long) block);
109 		return ERR_PTR(err);
110 	}
111 	dirent = (struct ext4_dir_entry *) bh->b_data;
112 	/* Determine whether or not we have an index block */
113 	if (is_dx(inode)) {
114 		if (block == 0)
115 			is_dx_block = 1;
116 		else if (ext4_rec_len_from_disk(dirent->rec_len,
117 						inode->i_sb->s_blocksize) ==
118 			 inode->i_sb->s_blocksize)
119 			is_dx_block = 1;
120 	}
121 	if (!is_dx_block && type == INDEX) {
122 		ext4_error_inode(inode, __func__, line, block,
123 		       "directory leaf block found instead of index block");
124 		return ERR_PTR(-EIO);
125 	}
126 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
127 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
128 	    buffer_verified(bh))
129 		return bh;
130 
131 	/*
132 	 * An empty leaf block can get mistaken for a index block; for
133 	 * this reason, we can only check the index checksum when the
134 	 * caller is sure it should be an index block.
135 	 */
136 	if (is_dx_block && type == INDEX) {
137 		if (ext4_dx_csum_verify(inode, dirent))
138 			set_buffer_verified(bh);
139 		else {
140 			ext4_error_inode(inode, __func__, line, block,
141 				"Directory index failed checksum");
142 			brelse(bh);
143 			return ERR_PTR(-EIO);
144 		}
145 	}
146 	if (!is_dx_block) {
147 		if (ext4_dirent_csum_verify(inode, dirent))
148 			set_buffer_verified(bh);
149 		else {
150 			ext4_error_inode(inode, __func__, line, block,
151 				"Directory block failed checksum");
152 			brelse(bh);
153 			return ERR_PTR(-EIO);
154 		}
155 	}
156 	return bh;
157 }
158 
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162 
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168 
169 struct fake_dirent
170 {
171 	__le32 inode;
172 	__le16 rec_len;
173 	u8 name_len;
174 	u8 file_type;
175 };
176 
177 struct dx_countlimit
178 {
179 	__le16 limit;
180 	__le16 count;
181 };
182 
183 struct dx_entry
184 {
185 	__le32 hash;
186 	__le32 block;
187 };
188 
189 /*
190  * dx_root_info is laid out so that if it should somehow get overlaid by a
191  * dirent the two low bits of the hash version will be zero.  Therefore, the
192  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
193  */
194 
195 struct dx_root
196 {
197 	struct fake_dirent dot;
198 	char dot_name[4];
199 	struct fake_dirent dotdot;
200 	char dotdot_name[4];
201 	struct dx_root_info
202 	{
203 		__le32 reserved_zero;
204 		u8 hash_version;
205 		u8 info_length; /* 8 */
206 		u8 indirect_levels;
207 		u8 unused_flags;
208 	}
209 	info;
210 	struct dx_entry	entries[0];
211 };
212 
213 struct dx_node
214 {
215 	struct fake_dirent fake;
216 	struct dx_entry	entries[0];
217 };
218 
219 
220 struct dx_frame
221 {
222 	struct buffer_head *bh;
223 	struct dx_entry *entries;
224 	struct dx_entry *at;
225 };
226 
227 struct dx_map_entry
228 {
229 	u32 hash;
230 	u16 offs;
231 	u16 size;
232 };
233 
234 /*
235  * This goes at the end of each htree block.
236  */
237 struct dx_tail {
238 	u32 dt_reserved;
239 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
240 };
241 
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253 				 struct inode *dir,
254 				 struct dx_hash_info *hinfo,
255 				 struct dx_frame *frame,
256 				 int *err);
257 static void dx_release(struct dx_frame *frames);
258 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
259 		       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
260 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
261 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
262 		struct dx_map_entry *offsets, int count, unsigned blocksize);
263 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
264 static void dx_insert_block(struct dx_frame *frame,
265 					u32 hash, ext4_lblk_t block);
266 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
267 				 struct dx_frame *frame,
268 				 struct dx_frame *frames,
269 				 __u32 *start_hash);
270 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
271 		const struct qstr *d_name,
272 		struct ext4_dir_entry_2 **res_dir,
273 		int *err);
274 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
275 			     struct inode *inode);
276 
277 /* checksumming functions */
278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
279 			    unsigned int blocksize)
280 {
281 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
282 	t->det_rec_len = ext4_rec_len_to_disk(
283 			sizeof(struct ext4_dir_entry_tail), blocksize);
284 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
285 }
286 
287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
289 						   struct ext4_dir_entry *de)
290 {
291 	struct ext4_dir_entry_tail *t;
292 
293 #ifdef PARANOID
294 	struct ext4_dir_entry *d, *top;
295 
296 	d = de;
297 	top = (struct ext4_dir_entry *)(((void *)de) +
298 		(EXT4_BLOCK_SIZE(inode->i_sb) -
299 		sizeof(struct ext4_dir_entry_tail)));
300 	while (d < top && d->rec_len)
301 		d = (struct ext4_dir_entry *)(((void *)d) +
302 		    le16_to_cpu(d->rec_len));
303 
304 	if (d != top)
305 		return NULL;
306 
307 	t = (struct ext4_dir_entry_tail *)d;
308 #else
309 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
310 #endif
311 
312 	if (t->det_reserved_zero1 ||
313 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
314 	    t->det_reserved_zero2 ||
315 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
316 		return NULL;
317 
318 	return t;
319 }
320 
321 static __le32 ext4_dirent_csum(struct inode *inode,
322 			       struct ext4_dir_entry *dirent, int size)
323 {
324 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
325 	struct ext4_inode_info *ei = EXT4_I(inode);
326 	__u32 csum;
327 
328 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
329 	return cpu_to_le32(csum);
330 }
331 
332 static void warn_no_space_for_csum(struct inode *inode)
333 {
334 	ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
335 		     "checksum.  Please run e2fsck -D.", inode->i_ino);
336 }
337 
338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
339 {
340 	struct ext4_dir_entry_tail *t;
341 
342 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
343 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
344 		return 1;
345 
346 	t = get_dirent_tail(inode, dirent);
347 	if (!t) {
348 		warn_no_space_for_csum(inode);
349 		return 0;
350 	}
351 
352 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
353 						(void *)t - (void *)dirent))
354 		return 0;
355 
356 	return 1;
357 }
358 
359 static void ext4_dirent_csum_set(struct inode *inode,
360 				 struct ext4_dir_entry *dirent)
361 {
362 	struct ext4_dir_entry_tail *t;
363 
364 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
365 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
366 		return;
367 
368 	t = get_dirent_tail(inode, dirent);
369 	if (!t) {
370 		warn_no_space_for_csum(inode);
371 		return;
372 	}
373 
374 	t->det_checksum = ext4_dirent_csum(inode, dirent,
375 					   (void *)t - (void *)dirent);
376 }
377 
378 int ext4_handle_dirty_dirent_node(handle_t *handle,
379 				  struct inode *inode,
380 				  struct buffer_head *bh)
381 {
382 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
383 	return ext4_handle_dirty_metadata(handle, inode, bh);
384 }
385 
386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
387 					       struct ext4_dir_entry *dirent,
388 					       int *offset)
389 {
390 	struct ext4_dir_entry *dp;
391 	struct dx_root_info *root;
392 	int count_offset;
393 
394 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
395 		count_offset = 8;
396 	else if (le16_to_cpu(dirent->rec_len) == 12) {
397 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
398 		if (le16_to_cpu(dp->rec_len) !=
399 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
400 			return NULL;
401 		root = (struct dx_root_info *)(((void *)dp + 12));
402 		if (root->reserved_zero ||
403 		    root->info_length != sizeof(struct dx_root_info))
404 			return NULL;
405 		count_offset = 32;
406 	} else
407 		return NULL;
408 
409 	if (offset)
410 		*offset = count_offset;
411 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
412 }
413 
414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
415 			   int count_offset, int count, struct dx_tail *t)
416 {
417 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
418 	struct ext4_inode_info *ei = EXT4_I(inode);
419 	__u32 csum;
420 	__le32 save_csum;
421 	int size;
422 
423 	size = count_offset + (count * sizeof(struct dx_entry));
424 	save_csum = t->dt_checksum;
425 	t->dt_checksum = 0;
426 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
427 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
428 	t->dt_checksum = save_csum;
429 
430 	return cpu_to_le32(csum);
431 }
432 
433 static int ext4_dx_csum_verify(struct inode *inode,
434 			       struct ext4_dir_entry *dirent)
435 {
436 	struct dx_countlimit *c;
437 	struct dx_tail *t;
438 	int count_offset, limit, count;
439 
440 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
441 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
442 		return 1;
443 
444 	c = get_dx_countlimit(inode, dirent, &count_offset);
445 	if (!c) {
446 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
447 		return 1;
448 	}
449 	limit = le16_to_cpu(c->limit);
450 	count = le16_to_cpu(c->count);
451 	if (count_offset + (limit * sizeof(struct dx_entry)) >
452 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
453 		warn_no_space_for_csum(inode);
454 		return 1;
455 	}
456 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
457 
458 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
459 					    count, t))
460 		return 0;
461 	return 1;
462 }
463 
464 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
465 {
466 	struct dx_countlimit *c;
467 	struct dx_tail *t;
468 	int count_offset, limit, count;
469 
470 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
471 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
472 		return;
473 
474 	c = get_dx_countlimit(inode, dirent, &count_offset);
475 	if (!c) {
476 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
477 		return;
478 	}
479 	limit = le16_to_cpu(c->limit);
480 	count = le16_to_cpu(c->count);
481 	if (count_offset + (limit * sizeof(struct dx_entry)) >
482 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 		warn_no_space_for_csum(inode);
484 		return;
485 	}
486 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487 
488 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490 
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 					    struct inode *inode,
493 					    struct buffer_head *bh)
494 {
495 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 	return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498 
499 /*
500  * p is at least 6 bytes before the end of page
501  */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 	return (struct ext4_dir_entry_2 *)((char *)p +
506 		ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508 
509 /*
510  * Future: use high four bits of block for coalesce-on-delete flags
511  * Mask them off for now.
512  */
513 
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 	return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518 
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 	entry->block = cpu_to_le32(value);
522 }
523 
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 	return le32_to_cpu(entry->hash);
527 }
528 
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 	entry->hash = cpu_to_le32(value);
532 }
533 
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538 
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543 
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548 
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553 
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 		EXT4_DIR_REC_LEN(2) - infosize;
558 
559 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
560 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
561 		entry_space -= sizeof(struct dx_tail);
562 	return entry_space / sizeof(struct dx_entry);
563 }
564 
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 
569 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
570 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
571 		entry_space -= sizeof(struct dx_tail);
572 	return entry_space / sizeof(struct dx_entry);
573 }
574 
575 /*
576  * Debug
577  */
578 #ifdef DX_DEBUG
579 static void dx_show_index(char * label, struct dx_entry *entries)
580 {
581 	int i, n = dx_get_count (entries);
582 	printk(KERN_DEBUG "%s index ", label);
583 	for (i = 0; i < n; i++) {
584 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
585 				0, (unsigned long)dx_get_block(entries + i));
586 	}
587 	printk("\n");
588 }
589 
590 struct stats
591 {
592 	unsigned names;
593 	unsigned space;
594 	unsigned bcount;
595 };
596 
597 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
598 				 int size, int show_names)
599 {
600 	unsigned names = 0, space = 0;
601 	char *base = (char *) de;
602 	struct dx_hash_info h = *hinfo;
603 
604 	printk("names: ");
605 	while ((char *) de < base + size)
606 	{
607 		if (de->inode)
608 		{
609 			if (show_names)
610 			{
611 				int len = de->name_len;
612 				char *name = de->name;
613 				while (len--) printk("%c", *name++);
614 				ext4fs_dirhash(de->name, de->name_len, &h);
615 				printk(":%x.%u ", h.hash,
616 				       (unsigned) ((char *) de - base));
617 			}
618 			space += EXT4_DIR_REC_LEN(de->name_len);
619 			names++;
620 		}
621 		de = ext4_next_entry(de, size);
622 	}
623 	printk("(%i)\n", names);
624 	return (struct stats) { names, space, 1 };
625 }
626 
627 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
628 			     struct dx_entry *entries, int levels)
629 {
630 	unsigned blocksize = dir->i_sb->s_blocksize;
631 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
632 	unsigned bcount = 0;
633 	struct buffer_head *bh;
634 	int err;
635 	printk("%i indexed blocks...\n", count);
636 	for (i = 0; i < count; i++, entries++)
637 	{
638 		ext4_lblk_t block = dx_get_block(entries);
639 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
640 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
641 		struct stats stats;
642 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
643 		if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
644 		stats = levels?
645 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
646 		   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
647 		names += stats.names;
648 		space += stats.space;
649 		bcount += stats.bcount;
650 		brelse(bh);
651 	}
652 	if (bcount)
653 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
654 		       levels ? "" : "   ", names, space/bcount,
655 		       (space/bcount)*100/blocksize);
656 	return (struct stats) { names, space, bcount};
657 }
658 #endif /* DX_DEBUG */
659 
660 /*
661  * Probe for a directory leaf block to search.
662  *
663  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
664  * error in the directory index, and the caller should fall back to
665  * searching the directory normally.  The callers of dx_probe **MUST**
666  * check for this error code, and make sure it never gets reflected
667  * back to userspace.
668  */
669 static struct dx_frame *
670 dx_probe(const struct qstr *d_name, struct inode *dir,
671 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
672 {
673 	unsigned count, indirect;
674 	struct dx_entry *at, *entries, *p, *q, *m;
675 	struct dx_root *root;
676 	struct buffer_head *bh;
677 	struct dx_frame *frame = frame_in;
678 	u32 hash;
679 
680 	frame->bh = NULL;
681 	bh = ext4_read_dirblock(dir, 0, INDEX);
682 	if (IS_ERR(bh)) {
683 		*err = PTR_ERR(bh);
684 		goto fail;
685 	}
686 	root = (struct dx_root *) bh->b_data;
687 	if (root->info.hash_version != DX_HASH_TEA &&
688 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
689 	    root->info.hash_version != DX_HASH_LEGACY) {
690 		ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
691 			     root->info.hash_version);
692 		brelse(bh);
693 		*err = ERR_BAD_DX_DIR;
694 		goto fail;
695 	}
696 	hinfo->hash_version = root->info.hash_version;
697 	if (hinfo->hash_version <= DX_HASH_TEA)
698 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
699 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
700 	if (d_name)
701 		ext4fs_dirhash(d_name->name, d_name->len, hinfo);
702 	hash = hinfo->hash;
703 
704 	if (root->info.unused_flags & 1) {
705 		ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
706 			     root->info.unused_flags);
707 		brelse(bh);
708 		*err = ERR_BAD_DX_DIR;
709 		goto fail;
710 	}
711 
712 	if ((indirect = root->info.indirect_levels) > 1) {
713 		ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
714 			     root->info.indirect_levels);
715 		brelse(bh);
716 		*err = ERR_BAD_DX_DIR;
717 		goto fail;
718 	}
719 
720 	entries = (struct dx_entry *) (((char *)&root->info) +
721 				       root->info.info_length);
722 
723 	if (dx_get_limit(entries) != dx_root_limit(dir,
724 						   root->info.info_length)) {
725 		ext4_warning(dir->i_sb, "dx entry: limit != root limit");
726 		brelse(bh);
727 		*err = ERR_BAD_DX_DIR;
728 		goto fail;
729 	}
730 
731 	dxtrace(printk("Look up %x", hash));
732 	while (1)
733 	{
734 		count = dx_get_count(entries);
735 		if (!count || count > dx_get_limit(entries)) {
736 			ext4_warning(dir->i_sb,
737 				     "dx entry: no count or count > limit");
738 			brelse(bh);
739 			*err = ERR_BAD_DX_DIR;
740 			goto fail2;
741 		}
742 
743 		p = entries + 1;
744 		q = entries + count - 1;
745 		while (p <= q)
746 		{
747 			m = p + (q - p)/2;
748 			dxtrace(printk("."));
749 			if (dx_get_hash(m) > hash)
750 				q = m - 1;
751 			else
752 				p = m + 1;
753 		}
754 
755 		if (0) // linear search cross check
756 		{
757 			unsigned n = count - 1;
758 			at = entries;
759 			while (n--)
760 			{
761 				dxtrace(printk(","));
762 				if (dx_get_hash(++at) > hash)
763 				{
764 					at--;
765 					break;
766 				}
767 			}
768 			assert (at == p - 1);
769 		}
770 
771 		at = p - 1;
772 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
773 		frame->bh = bh;
774 		frame->entries = entries;
775 		frame->at = at;
776 		if (!indirect--) return frame;
777 		bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
778 		if (IS_ERR(bh)) {
779 			*err = PTR_ERR(bh);
780 			goto fail2;
781 		}
782 		entries = ((struct dx_node *) bh->b_data)->entries;
783 
784 		if (dx_get_limit(entries) != dx_node_limit (dir)) {
785 			ext4_warning(dir->i_sb,
786 				     "dx entry: limit != node limit");
787 			brelse(bh);
788 			*err = ERR_BAD_DX_DIR;
789 			goto fail2;
790 		}
791 		frame++;
792 		frame->bh = NULL;
793 	}
794 fail2:
795 	while (frame >= frame_in) {
796 		brelse(frame->bh);
797 		frame--;
798 	}
799 fail:
800 	if (*err == ERR_BAD_DX_DIR)
801 		ext4_warning(dir->i_sb,
802 			     "Corrupt dir inode %lu, running e2fsck is "
803 			     "recommended.", dir->i_ino);
804 	return NULL;
805 }
806 
807 static void dx_release (struct dx_frame *frames)
808 {
809 	if (frames[0].bh == NULL)
810 		return;
811 
812 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
813 		brelse(frames[1].bh);
814 	brelse(frames[0].bh);
815 }
816 
817 /*
818  * This function increments the frame pointer to search the next leaf
819  * block, and reads in the necessary intervening nodes if the search
820  * should be necessary.  Whether or not the search is necessary is
821  * controlled by the hash parameter.  If the hash value is even, then
822  * the search is only continued if the next block starts with that
823  * hash value.  This is used if we are searching for a specific file.
824  *
825  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
826  *
827  * This function returns 1 if the caller should continue to search,
828  * or 0 if it should not.  If there is an error reading one of the
829  * index blocks, it will a negative error code.
830  *
831  * If start_hash is non-null, it will be filled in with the starting
832  * hash of the next page.
833  */
834 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
835 				 struct dx_frame *frame,
836 				 struct dx_frame *frames,
837 				 __u32 *start_hash)
838 {
839 	struct dx_frame *p;
840 	struct buffer_head *bh;
841 	int num_frames = 0;
842 	__u32 bhash;
843 
844 	p = frame;
845 	/*
846 	 * Find the next leaf page by incrementing the frame pointer.
847 	 * If we run out of entries in the interior node, loop around and
848 	 * increment pointer in the parent node.  When we break out of
849 	 * this loop, num_frames indicates the number of interior
850 	 * nodes need to be read.
851 	 */
852 	while (1) {
853 		if (++(p->at) < p->entries + dx_get_count(p->entries))
854 			break;
855 		if (p == frames)
856 			return 0;
857 		num_frames++;
858 		p--;
859 	}
860 
861 	/*
862 	 * If the hash is 1, then continue only if the next page has a
863 	 * continuation hash of any value.  This is used for readdir
864 	 * handling.  Otherwise, check to see if the hash matches the
865 	 * desired contiuation hash.  If it doesn't, return since
866 	 * there's no point to read in the successive index pages.
867 	 */
868 	bhash = dx_get_hash(p->at);
869 	if (start_hash)
870 		*start_hash = bhash;
871 	if ((hash & 1) == 0) {
872 		if ((bhash & ~1) != hash)
873 			return 0;
874 	}
875 	/*
876 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
877 	 * block so no check is necessary
878 	 */
879 	while (num_frames--) {
880 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
881 		if (IS_ERR(bh))
882 			return PTR_ERR(bh);
883 		p++;
884 		brelse(p->bh);
885 		p->bh = bh;
886 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
887 	}
888 	return 1;
889 }
890 
891 
892 /*
893  * This function fills a red-black tree with information from a
894  * directory block.  It returns the number directory entries loaded
895  * into the tree.  If there is an error it is returned in err.
896  */
897 static int htree_dirblock_to_tree(struct file *dir_file,
898 				  struct inode *dir, ext4_lblk_t block,
899 				  struct dx_hash_info *hinfo,
900 				  __u32 start_hash, __u32 start_minor_hash)
901 {
902 	struct buffer_head *bh;
903 	struct ext4_dir_entry_2 *de, *top;
904 	int err = 0, count = 0;
905 
906 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
907 							(unsigned long)block));
908 	bh = ext4_read_dirblock(dir, block, DIRENT);
909 	if (IS_ERR(bh))
910 		return PTR_ERR(bh);
911 
912 	de = (struct ext4_dir_entry_2 *) bh->b_data;
913 	top = (struct ext4_dir_entry_2 *) ((char *) de +
914 					   dir->i_sb->s_blocksize -
915 					   EXT4_DIR_REC_LEN(0));
916 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
917 		if (ext4_check_dir_entry(dir, NULL, de, bh,
918 				bh->b_data, bh->b_size,
919 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
920 					 + ((char *)de - bh->b_data))) {
921 			/* On error, skip the f_pos to the next block. */
922 			dir_file->f_pos = (dir_file->f_pos |
923 					(dir->i_sb->s_blocksize - 1)) + 1;
924 			brelse(bh);
925 			return count;
926 		}
927 		ext4fs_dirhash(de->name, de->name_len, hinfo);
928 		if ((hinfo->hash < start_hash) ||
929 		    ((hinfo->hash == start_hash) &&
930 		     (hinfo->minor_hash < start_minor_hash)))
931 			continue;
932 		if (de->inode == 0)
933 			continue;
934 		if ((err = ext4_htree_store_dirent(dir_file,
935 				   hinfo->hash, hinfo->minor_hash, de)) != 0) {
936 			brelse(bh);
937 			return err;
938 		}
939 		count++;
940 	}
941 	brelse(bh);
942 	return count;
943 }
944 
945 
946 /*
947  * This function fills a red-black tree with information from a
948  * directory.  We start scanning the directory in hash order, starting
949  * at start_hash and start_minor_hash.
950  *
951  * This function returns the number of entries inserted into the tree,
952  * or a negative error code.
953  */
954 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
955 			 __u32 start_minor_hash, __u32 *next_hash)
956 {
957 	struct dx_hash_info hinfo;
958 	struct ext4_dir_entry_2 *de;
959 	struct dx_frame frames[2], *frame;
960 	struct inode *dir;
961 	ext4_lblk_t block;
962 	int count = 0;
963 	int ret, err;
964 	__u32 hashval;
965 
966 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
967 		       start_hash, start_minor_hash));
968 	dir = file_inode(dir_file);
969 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
970 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
971 		if (hinfo.hash_version <= DX_HASH_TEA)
972 			hinfo.hash_version +=
973 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
974 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
975 		if (ext4_has_inline_data(dir)) {
976 			int has_inline_data = 1;
977 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
978 							&hinfo, start_hash,
979 							start_minor_hash,
980 							&has_inline_data);
981 			if (has_inline_data) {
982 				*next_hash = ~0;
983 				return count;
984 			}
985 		}
986 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
987 					       start_hash, start_minor_hash);
988 		*next_hash = ~0;
989 		return count;
990 	}
991 	hinfo.hash = start_hash;
992 	hinfo.minor_hash = 0;
993 	frame = dx_probe(NULL, dir, &hinfo, frames, &err);
994 	if (!frame)
995 		return err;
996 
997 	/* Add '.' and '..' from the htree header */
998 	if (!start_hash && !start_minor_hash) {
999 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1000 		if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
1001 			goto errout;
1002 		count++;
1003 	}
1004 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1005 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1006 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1007 		if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1008 			goto errout;
1009 		count++;
1010 	}
1011 
1012 	while (1) {
1013 		block = dx_get_block(frame->at);
1014 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1015 					     start_hash, start_minor_hash);
1016 		if (ret < 0) {
1017 			err = ret;
1018 			goto errout;
1019 		}
1020 		count += ret;
1021 		hashval = ~0;
1022 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1023 					    frame, frames, &hashval);
1024 		*next_hash = hashval;
1025 		if (ret < 0) {
1026 			err = ret;
1027 			goto errout;
1028 		}
1029 		/*
1030 		 * Stop if:  (a) there are no more entries, or
1031 		 * (b) we have inserted at least one entry and the
1032 		 * next hash value is not a continuation
1033 		 */
1034 		if ((ret == 0) ||
1035 		    (count && ((hashval & 1) == 0)))
1036 			break;
1037 	}
1038 	dx_release(frames);
1039 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1040 		       "next hash: %x\n", count, *next_hash));
1041 	return count;
1042 errout:
1043 	dx_release(frames);
1044 	return (err);
1045 }
1046 
1047 static inline int search_dirblock(struct buffer_head *bh,
1048 				  struct inode *dir,
1049 				  const struct qstr *d_name,
1050 				  unsigned int offset,
1051 				  struct ext4_dir_entry_2 **res_dir)
1052 {
1053 	return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1054 			  d_name, offset, res_dir);
1055 }
1056 
1057 /*
1058  * Directory block splitting, compacting
1059  */
1060 
1061 /*
1062  * Create map of hash values, offsets, and sizes, stored at end of block.
1063  * Returns number of entries mapped.
1064  */
1065 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1066 		       struct dx_hash_info *hinfo,
1067 		       struct dx_map_entry *map_tail)
1068 {
1069 	int count = 0;
1070 	char *base = (char *) de;
1071 	struct dx_hash_info h = *hinfo;
1072 
1073 	while ((char *) de < base + blocksize) {
1074 		if (de->name_len && de->inode) {
1075 			ext4fs_dirhash(de->name, de->name_len, &h);
1076 			map_tail--;
1077 			map_tail->hash = h.hash;
1078 			map_tail->offs = ((char *) de - base)>>2;
1079 			map_tail->size = le16_to_cpu(de->rec_len);
1080 			count++;
1081 			cond_resched();
1082 		}
1083 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1084 		de = ext4_next_entry(de, blocksize);
1085 	}
1086 	return count;
1087 }
1088 
1089 /* Sort map by hash value */
1090 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1091 {
1092 	struct dx_map_entry *p, *q, *top = map + count - 1;
1093 	int more;
1094 	/* Combsort until bubble sort doesn't suck */
1095 	while (count > 2) {
1096 		count = count*10/13;
1097 		if (count - 9 < 2) /* 9, 10 -> 11 */
1098 			count = 11;
1099 		for (p = top, q = p - count; q >= map; p--, q--)
1100 			if (p->hash < q->hash)
1101 				swap(*p, *q);
1102 	}
1103 	/* Garden variety bubble sort */
1104 	do {
1105 		more = 0;
1106 		q = top;
1107 		while (q-- > map) {
1108 			if (q[1].hash >= q[0].hash)
1109 				continue;
1110 			swap(*(q+1), *q);
1111 			more = 1;
1112 		}
1113 	} while(more);
1114 }
1115 
1116 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1117 {
1118 	struct dx_entry *entries = frame->entries;
1119 	struct dx_entry *old = frame->at, *new = old + 1;
1120 	int count = dx_get_count(entries);
1121 
1122 	assert(count < dx_get_limit(entries));
1123 	assert(old < entries + count);
1124 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1125 	dx_set_hash(new, hash);
1126 	dx_set_block(new, block);
1127 	dx_set_count(entries, count + 1);
1128 }
1129 
1130 /*
1131  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1132  *
1133  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1134  * `de != NULL' is guaranteed by caller.
1135  */
1136 static inline int ext4_match (int len, const char * const name,
1137 			      struct ext4_dir_entry_2 * de)
1138 {
1139 	if (len != de->name_len)
1140 		return 0;
1141 	if (!de->inode)
1142 		return 0;
1143 	return !memcmp(name, de->name, len);
1144 }
1145 
1146 /*
1147  * Returns 0 if not found, -1 on failure, and 1 on success
1148  */
1149 int search_dir(struct buffer_head *bh,
1150 	       char *search_buf,
1151 	       int buf_size,
1152 	       struct inode *dir,
1153 	       const struct qstr *d_name,
1154 	       unsigned int offset,
1155 	       struct ext4_dir_entry_2 **res_dir)
1156 {
1157 	struct ext4_dir_entry_2 * de;
1158 	char * dlimit;
1159 	int de_len;
1160 	const char *name = d_name->name;
1161 	int namelen = d_name->len;
1162 
1163 	de = (struct ext4_dir_entry_2 *)search_buf;
1164 	dlimit = search_buf + buf_size;
1165 	while ((char *) de < dlimit) {
1166 		/* this code is executed quadratically often */
1167 		/* do minimal checking `by hand' */
1168 
1169 		if ((char *) de + namelen <= dlimit &&
1170 		    ext4_match (namelen, name, de)) {
1171 			/* found a match - just to be sure, do a full check */
1172 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1173 						 bh->b_size, offset))
1174 				return -1;
1175 			*res_dir = de;
1176 			return 1;
1177 		}
1178 		/* prevent looping on a bad block */
1179 		de_len = ext4_rec_len_from_disk(de->rec_len,
1180 						dir->i_sb->s_blocksize);
1181 		if (de_len <= 0)
1182 			return -1;
1183 		offset += de_len;
1184 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1185 	}
1186 	return 0;
1187 }
1188 
1189 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1190 			       struct ext4_dir_entry *de)
1191 {
1192 	struct super_block *sb = dir->i_sb;
1193 
1194 	if (!is_dx(dir))
1195 		return 0;
1196 	if (block == 0)
1197 		return 1;
1198 	if (de->inode == 0 &&
1199 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1200 			sb->s_blocksize)
1201 		return 1;
1202 	return 0;
1203 }
1204 
1205 /*
1206  *	ext4_find_entry()
1207  *
1208  * finds an entry in the specified directory with the wanted name. It
1209  * returns the cache buffer in which the entry was found, and the entry
1210  * itself (as a parameter - res_dir). It does NOT read the inode of the
1211  * entry - you'll have to do that yourself if you want to.
1212  *
1213  * The returned buffer_head has ->b_count elevated.  The caller is expected
1214  * to brelse() it when appropriate.
1215  */
1216 static struct buffer_head * ext4_find_entry (struct inode *dir,
1217 					const struct qstr *d_name,
1218 					struct ext4_dir_entry_2 **res_dir,
1219 					int *inlined)
1220 {
1221 	struct super_block *sb;
1222 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1223 	struct buffer_head *bh, *ret = NULL;
1224 	ext4_lblk_t start, block, b;
1225 	const u8 *name = d_name->name;
1226 	int ra_max = 0;		/* Number of bh's in the readahead
1227 				   buffer, bh_use[] */
1228 	int ra_ptr = 0;		/* Current index into readahead
1229 				   buffer */
1230 	int num = 0;
1231 	ext4_lblk_t  nblocks;
1232 	int i, err;
1233 	int namelen;
1234 
1235 	*res_dir = NULL;
1236 	sb = dir->i_sb;
1237 	namelen = d_name->len;
1238 	if (namelen > EXT4_NAME_LEN)
1239 		return NULL;
1240 
1241 	if (ext4_has_inline_data(dir)) {
1242 		int has_inline_data = 1;
1243 		ret = ext4_find_inline_entry(dir, d_name, res_dir,
1244 					     &has_inline_data);
1245 		if (has_inline_data) {
1246 			if (inlined)
1247 				*inlined = 1;
1248 			return ret;
1249 		}
1250 	}
1251 
1252 	if ((namelen <= 2) && (name[0] == '.') &&
1253 	    (name[1] == '.' || name[1] == '\0')) {
1254 		/*
1255 		 * "." or ".." will only be in the first block
1256 		 * NFS may look up ".."; "." should be handled by the VFS
1257 		 */
1258 		block = start = 0;
1259 		nblocks = 1;
1260 		goto restart;
1261 	}
1262 	if (is_dx(dir)) {
1263 		bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1264 		/*
1265 		 * On success, or if the error was file not found,
1266 		 * return.  Otherwise, fall back to doing a search the
1267 		 * old fashioned way.
1268 		 */
1269 		if (bh || (err != ERR_BAD_DX_DIR))
1270 			return bh;
1271 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1272 			       "falling back\n"));
1273 	}
1274 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1275 	start = EXT4_I(dir)->i_dir_start_lookup;
1276 	if (start >= nblocks)
1277 		start = 0;
1278 	block = start;
1279 restart:
1280 	do {
1281 		/*
1282 		 * We deal with the read-ahead logic here.
1283 		 */
1284 		if (ra_ptr >= ra_max) {
1285 			/* Refill the readahead buffer */
1286 			ra_ptr = 0;
1287 			b = block;
1288 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1289 				/*
1290 				 * Terminate if we reach the end of the
1291 				 * directory and must wrap, or if our
1292 				 * search has finished at this block.
1293 				 */
1294 				if (b >= nblocks || (num && block == start)) {
1295 					bh_use[ra_max] = NULL;
1296 					break;
1297 				}
1298 				num++;
1299 				bh = ext4_getblk(NULL, dir, b++, 0, &err);
1300 				bh_use[ra_max] = bh;
1301 				if (bh)
1302 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1303 						    1, &bh);
1304 			}
1305 		}
1306 		if ((bh = bh_use[ra_ptr++]) == NULL)
1307 			goto next;
1308 		wait_on_buffer(bh);
1309 		if (!buffer_uptodate(bh)) {
1310 			/* read error, skip block & hope for the best */
1311 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1312 					 (unsigned long) block);
1313 			brelse(bh);
1314 			goto next;
1315 		}
1316 		if (!buffer_verified(bh) &&
1317 		    !is_dx_internal_node(dir, block,
1318 					 (struct ext4_dir_entry *)bh->b_data) &&
1319 		    !ext4_dirent_csum_verify(dir,
1320 				(struct ext4_dir_entry *)bh->b_data)) {
1321 			EXT4_ERROR_INODE(dir, "checksumming directory "
1322 					 "block %lu", (unsigned long)block);
1323 			brelse(bh);
1324 			goto next;
1325 		}
1326 		set_buffer_verified(bh);
1327 		i = search_dirblock(bh, dir, d_name,
1328 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1329 		if (i == 1) {
1330 			EXT4_I(dir)->i_dir_start_lookup = block;
1331 			ret = bh;
1332 			goto cleanup_and_exit;
1333 		} else {
1334 			brelse(bh);
1335 			if (i < 0)
1336 				goto cleanup_and_exit;
1337 		}
1338 	next:
1339 		if (++block >= nblocks)
1340 			block = 0;
1341 	} while (block != start);
1342 
1343 	/*
1344 	 * If the directory has grown while we were searching, then
1345 	 * search the last part of the directory before giving up.
1346 	 */
1347 	block = nblocks;
1348 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1349 	if (block < nblocks) {
1350 		start = 0;
1351 		goto restart;
1352 	}
1353 
1354 cleanup_and_exit:
1355 	/* Clean up the read-ahead blocks */
1356 	for (; ra_ptr < ra_max; ra_ptr++)
1357 		brelse(bh_use[ra_ptr]);
1358 	return ret;
1359 }
1360 
1361 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1362 		       struct ext4_dir_entry_2 **res_dir, int *err)
1363 {
1364 	struct super_block * sb = dir->i_sb;
1365 	struct dx_hash_info	hinfo;
1366 	struct dx_frame frames[2], *frame;
1367 	struct buffer_head *bh;
1368 	ext4_lblk_t block;
1369 	int retval;
1370 
1371 	if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1372 		return NULL;
1373 	do {
1374 		block = dx_get_block(frame->at);
1375 		bh = ext4_read_dirblock(dir, block, DIRENT);
1376 		if (IS_ERR(bh)) {
1377 			*err = PTR_ERR(bh);
1378 			goto errout;
1379 		}
1380 		retval = search_dirblock(bh, dir, d_name,
1381 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1382 					 res_dir);
1383 		if (retval == 1) { 	/* Success! */
1384 			dx_release(frames);
1385 			return bh;
1386 		}
1387 		brelse(bh);
1388 		if (retval == -1) {
1389 			*err = ERR_BAD_DX_DIR;
1390 			goto errout;
1391 		}
1392 
1393 		/* Check to see if we should continue to search */
1394 		retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1395 					       frames, NULL);
1396 		if (retval < 0) {
1397 			ext4_warning(sb,
1398 			     "error reading index page in directory #%lu",
1399 			     dir->i_ino);
1400 			*err = retval;
1401 			goto errout;
1402 		}
1403 	} while (retval == 1);
1404 
1405 	*err = -ENOENT;
1406 errout:
1407 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1408 	dx_release (frames);
1409 	return NULL;
1410 }
1411 
1412 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1413 {
1414 	struct inode *inode;
1415 	struct ext4_dir_entry_2 *de;
1416 	struct buffer_head *bh;
1417 
1418 	if (dentry->d_name.len > EXT4_NAME_LEN)
1419 		return ERR_PTR(-ENAMETOOLONG);
1420 
1421 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1422 	inode = NULL;
1423 	if (bh) {
1424 		__u32 ino = le32_to_cpu(de->inode);
1425 		brelse(bh);
1426 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1427 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1428 			return ERR_PTR(-EIO);
1429 		}
1430 		if (unlikely(ino == dir->i_ino)) {
1431 			EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1432 					 dentry->d_name.len,
1433 					 dentry->d_name.name);
1434 			return ERR_PTR(-EIO);
1435 		}
1436 		inode = ext4_iget(dir->i_sb, ino);
1437 		if (inode == ERR_PTR(-ESTALE)) {
1438 			EXT4_ERROR_INODE(dir,
1439 					 "deleted inode referenced: %u",
1440 					 ino);
1441 			return ERR_PTR(-EIO);
1442 		}
1443 	}
1444 	return d_splice_alias(inode, dentry);
1445 }
1446 
1447 
1448 struct dentry *ext4_get_parent(struct dentry *child)
1449 {
1450 	__u32 ino;
1451 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1452 	struct ext4_dir_entry_2 * de;
1453 	struct buffer_head *bh;
1454 
1455 	bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1456 	if (!bh)
1457 		return ERR_PTR(-ENOENT);
1458 	ino = le32_to_cpu(de->inode);
1459 	brelse(bh);
1460 
1461 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1462 		EXT4_ERROR_INODE(child->d_inode,
1463 				 "bad parent inode number: %u", ino);
1464 		return ERR_PTR(-EIO);
1465 	}
1466 
1467 	return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1468 }
1469 
1470 /*
1471  * Move count entries from end of map between two memory locations.
1472  * Returns pointer to last entry moved.
1473  */
1474 static struct ext4_dir_entry_2 *
1475 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1476 		unsigned blocksize)
1477 {
1478 	unsigned rec_len = 0;
1479 
1480 	while (count--) {
1481 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1482 						(from + (map->offs<<2));
1483 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1484 		memcpy (to, de, rec_len);
1485 		((struct ext4_dir_entry_2 *) to)->rec_len =
1486 				ext4_rec_len_to_disk(rec_len, blocksize);
1487 		de->inode = 0;
1488 		map++;
1489 		to += rec_len;
1490 	}
1491 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1492 }
1493 
1494 /*
1495  * Compact each dir entry in the range to the minimal rec_len.
1496  * Returns pointer to last entry in range.
1497  */
1498 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1499 {
1500 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1501 	unsigned rec_len = 0;
1502 
1503 	prev = to = de;
1504 	while ((char*)de < base + blocksize) {
1505 		next = ext4_next_entry(de, blocksize);
1506 		if (de->inode && de->name_len) {
1507 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1508 			if (de > to)
1509 				memmove(to, de, rec_len);
1510 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1511 			prev = to;
1512 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1513 		}
1514 		de = next;
1515 	}
1516 	return prev;
1517 }
1518 
1519 /*
1520  * Split a full leaf block to make room for a new dir entry.
1521  * Allocate a new block, and move entries so that they are approx. equally full.
1522  * Returns pointer to de in block into which the new entry will be inserted.
1523  */
1524 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1525 			struct buffer_head **bh,struct dx_frame *frame,
1526 			struct dx_hash_info *hinfo, int *error)
1527 {
1528 	unsigned blocksize = dir->i_sb->s_blocksize;
1529 	unsigned count, continued;
1530 	struct buffer_head *bh2;
1531 	ext4_lblk_t newblock;
1532 	u32 hash2;
1533 	struct dx_map_entry *map;
1534 	char *data1 = (*bh)->b_data, *data2;
1535 	unsigned split, move, size;
1536 	struct ext4_dir_entry_2 *de = NULL, *de2;
1537 	struct ext4_dir_entry_tail *t;
1538 	int	csum_size = 0;
1539 	int	err = 0, i;
1540 
1541 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1542 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1543 		csum_size = sizeof(struct ext4_dir_entry_tail);
1544 
1545 	bh2 = ext4_append(handle, dir, &newblock);
1546 	if (IS_ERR(bh2)) {
1547 		brelse(*bh);
1548 		*bh = NULL;
1549 		*error = PTR_ERR(bh2);
1550 		return NULL;
1551 	}
1552 
1553 	BUFFER_TRACE(*bh, "get_write_access");
1554 	err = ext4_journal_get_write_access(handle, *bh);
1555 	if (err)
1556 		goto journal_error;
1557 
1558 	BUFFER_TRACE(frame->bh, "get_write_access");
1559 	err = ext4_journal_get_write_access(handle, frame->bh);
1560 	if (err)
1561 		goto journal_error;
1562 
1563 	data2 = bh2->b_data;
1564 
1565 	/* create map in the end of data2 block */
1566 	map = (struct dx_map_entry *) (data2 + blocksize);
1567 	count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1568 			     blocksize, hinfo, map);
1569 	map -= count;
1570 	dx_sort_map(map, count);
1571 	/* Split the existing block in the middle, size-wise */
1572 	size = 0;
1573 	move = 0;
1574 	for (i = count-1; i >= 0; i--) {
1575 		/* is more than half of this entry in 2nd half of the block? */
1576 		if (size + map[i].size/2 > blocksize/2)
1577 			break;
1578 		size += map[i].size;
1579 		move++;
1580 	}
1581 	/* map index at which we will split */
1582 	split = count - move;
1583 	hash2 = map[split].hash;
1584 	continued = hash2 == map[split - 1].hash;
1585 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1586 			(unsigned long)dx_get_block(frame->at),
1587 					hash2, split, count-split));
1588 
1589 	/* Fancy dance to stay within two buffers */
1590 	de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1591 	de = dx_pack_dirents(data1, blocksize);
1592 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1593 					   (char *) de,
1594 					   blocksize);
1595 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1596 					    (char *) de2,
1597 					    blocksize);
1598 	if (csum_size) {
1599 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1600 		initialize_dirent_tail(t, blocksize);
1601 
1602 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1603 		initialize_dirent_tail(t, blocksize);
1604 	}
1605 
1606 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1607 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1608 
1609 	/* Which block gets the new entry? */
1610 	if (hinfo->hash >= hash2)
1611 	{
1612 		swap(*bh, bh2);
1613 		de = de2;
1614 	}
1615 	dx_insert_block(frame, hash2 + continued, newblock);
1616 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1617 	if (err)
1618 		goto journal_error;
1619 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1620 	if (err)
1621 		goto journal_error;
1622 	brelse(bh2);
1623 	dxtrace(dx_show_index("frame", frame->entries));
1624 	return de;
1625 
1626 journal_error:
1627 	brelse(*bh);
1628 	brelse(bh2);
1629 	*bh = NULL;
1630 	ext4_std_error(dir->i_sb, err);
1631 	*error = err;
1632 	return NULL;
1633 }
1634 
1635 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1636 		      struct buffer_head *bh,
1637 		      void *buf, int buf_size,
1638 		      const char *name, int namelen,
1639 		      struct ext4_dir_entry_2 **dest_de)
1640 {
1641 	struct ext4_dir_entry_2 *de;
1642 	unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1643 	int nlen, rlen;
1644 	unsigned int offset = 0;
1645 	char *top;
1646 
1647 	de = (struct ext4_dir_entry_2 *)buf;
1648 	top = buf + buf_size - reclen;
1649 	while ((char *) de <= top) {
1650 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1651 					 buf, buf_size, offset))
1652 			return -EIO;
1653 		if (ext4_match(namelen, name, de))
1654 			return -EEXIST;
1655 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1656 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1657 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1658 			break;
1659 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1660 		offset += rlen;
1661 	}
1662 	if ((char *) de > top)
1663 		return -ENOSPC;
1664 
1665 	*dest_de = de;
1666 	return 0;
1667 }
1668 
1669 void ext4_insert_dentry(struct inode *inode,
1670 			struct ext4_dir_entry_2 *de,
1671 			int buf_size,
1672 			const char *name, int namelen)
1673 {
1674 
1675 	int nlen, rlen;
1676 
1677 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1678 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1679 	if (de->inode) {
1680 		struct ext4_dir_entry_2 *de1 =
1681 				(struct ext4_dir_entry_2 *)((char *)de + nlen);
1682 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1683 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1684 		de = de1;
1685 	}
1686 	de->file_type = EXT4_FT_UNKNOWN;
1687 	de->inode = cpu_to_le32(inode->i_ino);
1688 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1689 	de->name_len = namelen;
1690 	memcpy(de->name, name, namelen);
1691 }
1692 /*
1693  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1694  * it points to a directory entry which is guaranteed to be large
1695  * enough for new directory entry.  If de is NULL, then
1696  * add_dirent_to_buf will attempt search the directory block for
1697  * space.  It will return -ENOSPC if no space is available, and -EIO
1698  * and -EEXIST if directory entry already exists.
1699  */
1700 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1701 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1702 			     struct buffer_head *bh)
1703 {
1704 	struct inode	*dir = dentry->d_parent->d_inode;
1705 	const char	*name = dentry->d_name.name;
1706 	int		namelen = dentry->d_name.len;
1707 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1708 	int		csum_size = 0;
1709 	int		err;
1710 
1711 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1712 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1713 		csum_size = sizeof(struct ext4_dir_entry_tail);
1714 
1715 	if (!de) {
1716 		err = ext4_find_dest_de(dir, inode,
1717 					bh, bh->b_data, blocksize - csum_size,
1718 					name, namelen, &de);
1719 		if (err)
1720 			return err;
1721 	}
1722 	BUFFER_TRACE(bh, "get_write_access");
1723 	err = ext4_journal_get_write_access(handle, bh);
1724 	if (err) {
1725 		ext4_std_error(dir->i_sb, err);
1726 		return err;
1727 	}
1728 
1729 	/* By now the buffer is marked for journaling */
1730 	ext4_insert_dentry(inode, de, blocksize, name, namelen);
1731 
1732 	/*
1733 	 * XXX shouldn't update any times until successful
1734 	 * completion of syscall, but too many callers depend
1735 	 * on this.
1736 	 *
1737 	 * XXX similarly, too many callers depend on
1738 	 * ext4_new_inode() setting the times, but error
1739 	 * recovery deletes the inode, so the worst that can
1740 	 * happen is that the times are slightly out of date
1741 	 * and/or different from the directory change time.
1742 	 */
1743 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1744 	ext4_update_dx_flag(dir);
1745 	dir->i_version++;
1746 	ext4_mark_inode_dirty(handle, dir);
1747 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1748 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1749 	if (err)
1750 		ext4_std_error(dir->i_sb, err);
1751 	return 0;
1752 }
1753 
1754 /*
1755  * This converts a one block unindexed directory to a 3 block indexed
1756  * directory, and adds the dentry to the indexed directory.
1757  */
1758 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1759 			    struct inode *inode, struct buffer_head *bh)
1760 {
1761 	struct inode	*dir = dentry->d_parent->d_inode;
1762 	const char	*name = dentry->d_name.name;
1763 	int		namelen = dentry->d_name.len;
1764 	struct buffer_head *bh2;
1765 	struct dx_root	*root;
1766 	struct dx_frame	frames[2], *frame;
1767 	struct dx_entry *entries;
1768 	struct ext4_dir_entry_2	*de, *de2;
1769 	struct ext4_dir_entry_tail *t;
1770 	char		*data1, *top;
1771 	unsigned	len;
1772 	int		retval;
1773 	unsigned	blocksize;
1774 	struct dx_hash_info hinfo;
1775 	ext4_lblk_t  block;
1776 	struct fake_dirent *fde;
1777 	int		csum_size = 0;
1778 
1779 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1780 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1781 		csum_size = sizeof(struct ext4_dir_entry_tail);
1782 
1783 	blocksize =  dir->i_sb->s_blocksize;
1784 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1785 	retval = ext4_journal_get_write_access(handle, bh);
1786 	if (retval) {
1787 		ext4_std_error(dir->i_sb, retval);
1788 		brelse(bh);
1789 		return retval;
1790 	}
1791 	root = (struct dx_root *) bh->b_data;
1792 
1793 	/* The 0th block becomes the root, move the dirents out */
1794 	fde = &root->dotdot;
1795 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1796 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1797 	if ((char *) de >= (((char *) root) + blocksize)) {
1798 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1799 		brelse(bh);
1800 		return -EIO;
1801 	}
1802 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1803 
1804 	/* Allocate new block for the 0th block's dirents */
1805 	bh2 = ext4_append(handle, dir, &block);
1806 	if (IS_ERR(bh2)) {
1807 		brelse(bh);
1808 		return PTR_ERR(bh2);
1809 	}
1810 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1811 	data1 = bh2->b_data;
1812 
1813 	memcpy (data1, de, len);
1814 	de = (struct ext4_dir_entry_2 *) data1;
1815 	top = data1 + len;
1816 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1817 		de = de2;
1818 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1819 					   (char *) de,
1820 					   blocksize);
1821 
1822 	if (csum_size) {
1823 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1824 		initialize_dirent_tail(t, blocksize);
1825 	}
1826 
1827 	/* Initialize the root; the dot dirents already exist */
1828 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1829 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1830 					   blocksize);
1831 	memset (&root->info, 0, sizeof(root->info));
1832 	root->info.info_length = sizeof(root->info);
1833 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1834 	entries = root->entries;
1835 	dx_set_block(entries, 1);
1836 	dx_set_count(entries, 1);
1837 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1838 
1839 	/* Initialize as for dx_probe */
1840 	hinfo.hash_version = root->info.hash_version;
1841 	if (hinfo.hash_version <= DX_HASH_TEA)
1842 		hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1843 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1844 	ext4fs_dirhash(name, namelen, &hinfo);
1845 	frame = frames;
1846 	frame->entries = entries;
1847 	frame->at = entries;
1848 	frame->bh = bh;
1849 	bh = bh2;
1850 
1851 	ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1852 	ext4_handle_dirty_dirent_node(handle, dir, bh);
1853 
1854 	de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1855 	if (!de) {
1856 		/*
1857 		 * Even if the block split failed, we have to properly write
1858 		 * out all the changes we did so far. Otherwise we can end up
1859 		 * with corrupted filesystem.
1860 		 */
1861 		ext4_mark_inode_dirty(handle, dir);
1862 		dx_release(frames);
1863 		return retval;
1864 	}
1865 	dx_release(frames);
1866 
1867 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1868 	brelse(bh);
1869 	return retval;
1870 }
1871 
1872 /*
1873  *	ext4_add_entry()
1874  *
1875  * adds a file entry to the specified directory, using the same
1876  * semantics as ext4_find_entry(). It returns NULL if it failed.
1877  *
1878  * NOTE!! The inode part of 'de' is left at 0 - which means you
1879  * may not sleep between calling this and putting something into
1880  * the entry, as someone else might have used it while you slept.
1881  */
1882 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1883 			  struct inode *inode)
1884 {
1885 	struct inode *dir = dentry->d_parent->d_inode;
1886 	struct buffer_head *bh;
1887 	struct ext4_dir_entry_2 *de;
1888 	struct ext4_dir_entry_tail *t;
1889 	struct super_block *sb;
1890 	int	retval;
1891 	int	dx_fallback=0;
1892 	unsigned blocksize;
1893 	ext4_lblk_t block, blocks;
1894 	int	csum_size = 0;
1895 
1896 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1897 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1898 		csum_size = sizeof(struct ext4_dir_entry_tail);
1899 
1900 	sb = dir->i_sb;
1901 	blocksize = sb->s_blocksize;
1902 	if (!dentry->d_name.len)
1903 		return -EINVAL;
1904 
1905 	if (ext4_has_inline_data(dir)) {
1906 		retval = ext4_try_add_inline_entry(handle, dentry, inode);
1907 		if (retval < 0)
1908 			return retval;
1909 		if (retval == 1) {
1910 			retval = 0;
1911 			return retval;
1912 		}
1913 	}
1914 
1915 	if (is_dx(dir)) {
1916 		retval = ext4_dx_add_entry(handle, dentry, inode);
1917 		if (!retval || (retval != ERR_BAD_DX_DIR))
1918 			return retval;
1919 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1920 		dx_fallback++;
1921 		ext4_mark_inode_dirty(handle, dir);
1922 	}
1923 	blocks = dir->i_size >> sb->s_blocksize_bits;
1924 	for (block = 0; block < blocks; block++) {
1925 		bh = ext4_read_dirblock(dir, block, DIRENT);
1926 		if (IS_ERR(bh))
1927 			return PTR_ERR(bh);
1928 
1929 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1930 		if (retval != -ENOSPC) {
1931 			brelse(bh);
1932 			return retval;
1933 		}
1934 
1935 		if (blocks == 1 && !dx_fallback &&
1936 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1937 			return make_indexed_dir(handle, dentry, inode, bh);
1938 		brelse(bh);
1939 	}
1940 	bh = ext4_append(handle, dir, &block);
1941 	if (IS_ERR(bh))
1942 		return PTR_ERR(bh);
1943 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1944 	de->inode = 0;
1945 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1946 
1947 	if (csum_size) {
1948 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1949 		initialize_dirent_tail(t, blocksize);
1950 	}
1951 
1952 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1953 	brelse(bh);
1954 	if (retval == 0)
1955 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1956 	return retval;
1957 }
1958 
1959 /*
1960  * Returns 0 for success, or a negative error value
1961  */
1962 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1963 			     struct inode *inode)
1964 {
1965 	struct dx_frame frames[2], *frame;
1966 	struct dx_entry *entries, *at;
1967 	struct dx_hash_info hinfo;
1968 	struct buffer_head *bh;
1969 	struct inode *dir = dentry->d_parent->d_inode;
1970 	struct super_block *sb = dir->i_sb;
1971 	struct ext4_dir_entry_2 *de;
1972 	int err;
1973 
1974 	frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1975 	if (!frame)
1976 		return err;
1977 	entries = frame->entries;
1978 	at = frame->at;
1979 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1980 	if (IS_ERR(bh)) {
1981 		err = PTR_ERR(bh);
1982 		bh = NULL;
1983 		goto cleanup;
1984 	}
1985 
1986 	BUFFER_TRACE(bh, "get_write_access");
1987 	err = ext4_journal_get_write_access(handle, bh);
1988 	if (err)
1989 		goto journal_error;
1990 
1991 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1992 	if (err != -ENOSPC)
1993 		goto cleanup;
1994 
1995 	/* Block full, should compress but for now just split */
1996 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1997 		       dx_get_count(entries), dx_get_limit(entries)));
1998 	/* Need to split index? */
1999 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2000 		ext4_lblk_t newblock;
2001 		unsigned icount = dx_get_count(entries);
2002 		int levels = frame - frames;
2003 		struct dx_entry *entries2;
2004 		struct dx_node *node2;
2005 		struct buffer_head *bh2;
2006 
2007 		if (levels && (dx_get_count(frames->entries) ==
2008 			       dx_get_limit(frames->entries))) {
2009 			ext4_warning(sb, "Directory index full!");
2010 			err = -ENOSPC;
2011 			goto cleanup;
2012 		}
2013 		bh2 = ext4_append(handle, dir, &newblock);
2014 		if (IS_ERR(bh2)) {
2015 			err = PTR_ERR(bh2);
2016 			goto cleanup;
2017 		}
2018 		node2 = (struct dx_node *)(bh2->b_data);
2019 		entries2 = node2->entries;
2020 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2021 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2022 							   sb->s_blocksize);
2023 		BUFFER_TRACE(frame->bh, "get_write_access");
2024 		err = ext4_journal_get_write_access(handle, frame->bh);
2025 		if (err)
2026 			goto journal_error;
2027 		if (levels) {
2028 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2029 			unsigned hash2 = dx_get_hash(entries + icount1);
2030 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2031 				       icount1, icount2));
2032 
2033 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2034 			err = ext4_journal_get_write_access(handle,
2035 							     frames[0].bh);
2036 			if (err)
2037 				goto journal_error;
2038 
2039 			memcpy((char *) entries2, (char *) (entries + icount1),
2040 			       icount2 * sizeof(struct dx_entry));
2041 			dx_set_count(entries, icount1);
2042 			dx_set_count(entries2, icount2);
2043 			dx_set_limit(entries2, dx_node_limit(dir));
2044 
2045 			/* Which index block gets the new entry? */
2046 			if (at - entries >= icount1) {
2047 				frame->at = at = at - entries - icount1 + entries2;
2048 				frame->entries = entries = entries2;
2049 				swap(frame->bh, bh2);
2050 			}
2051 			dx_insert_block(frames + 0, hash2, newblock);
2052 			dxtrace(dx_show_index("node", frames[1].entries));
2053 			dxtrace(dx_show_index("node",
2054 			       ((struct dx_node *) bh2->b_data)->entries));
2055 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2056 			if (err)
2057 				goto journal_error;
2058 			brelse (bh2);
2059 		} else {
2060 			dxtrace(printk(KERN_DEBUG
2061 				       "Creating second level index...\n"));
2062 			memcpy((char *) entries2, (char *) entries,
2063 			       icount * sizeof(struct dx_entry));
2064 			dx_set_limit(entries2, dx_node_limit(dir));
2065 
2066 			/* Set up root */
2067 			dx_set_count(entries, 1);
2068 			dx_set_block(entries + 0, newblock);
2069 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2070 
2071 			/* Add new access path frame */
2072 			frame = frames + 1;
2073 			frame->at = at = at - entries + entries2;
2074 			frame->entries = entries = entries2;
2075 			frame->bh = bh2;
2076 			err = ext4_journal_get_write_access(handle,
2077 							     frame->bh);
2078 			if (err)
2079 				goto journal_error;
2080 		}
2081 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2082 		if (err) {
2083 			ext4_std_error(inode->i_sb, err);
2084 			goto cleanup;
2085 		}
2086 	}
2087 	de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2088 	if (!de)
2089 		goto cleanup;
2090 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2091 	goto cleanup;
2092 
2093 journal_error:
2094 	ext4_std_error(dir->i_sb, err);
2095 cleanup:
2096 	brelse(bh);
2097 	dx_release(frames);
2098 	return err;
2099 }
2100 
2101 /*
2102  * ext4_generic_delete_entry deletes a directory entry by merging it
2103  * with the previous entry
2104  */
2105 int ext4_generic_delete_entry(handle_t *handle,
2106 			      struct inode *dir,
2107 			      struct ext4_dir_entry_2 *de_del,
2108 			      struct buffer_head *bh,
2109 			      void *entry_buf,
2110 			      int buf_size,
2111 			      int csum_size)
2112 {
2113 	struct ext4_dir_entry_2 *de, *pde;
2114 	unsigned int blocksize = dir->i_sb->s_blocksize;
2115 	int i;
2116 
2117 	i = 0;
2118 	pde = NULL;
2119 	de = (struct ext4_dir_entry_2 *)entry_buf;
2120 	while (i < buf_size - csum_size) {
2121 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2122 					 bh->b_data, bh->b_size, i))
2123 			return -EIO;
2124 		if (de == de_del)  {
2125 			if (pde)
2126 				pde->rec_len = ext4_rec_len_to_disk(
2127 					ext4_rec_len_from_disk(pde->rec_len,
2128 							       blocksize) +
2129 					ext4_rec_len_from_disk(de->rec_len,
2130 							       blocksize),
2131 					blocksize);
2132 			else
2133 				de->inode = 0;
2134 			dir->i_version++;
2135 			return 0;
2136 		}
2137 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2138 		pde = de;
2139 		de = ext4_next_entry(de, blocksize);
2140 	}
2141 	return -ENOENT;
2142 }
2143 
2144 static int ext4_delete_entry(handle_t *handle,
2145 			     struct inode *dir,
2146 			     struct ext4_dir_entry_2 *de_del,
2147 			     struct buffer_head *bh)
2148 {
2149 	int err, csum_size = 0;
2150 
2151 	if (ext4_has_inline_data(dir)) {
2152 		int has_inline_data = 1;
2153 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2154 					       &has_inline_data);
2155 		if (has_inline_data)
2156 			return err;
2157 	}
2158 
2159 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2160 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2161 		csum_size = sizeof(struct ext4_dir_entry_tail);
2162 
2163 	BUFFER_TRACE(bh, "get_write_access");
2164 	err = ext4_journal_get_write_access(handle, bh);
2165 	if (unlikely(err))
2166 		goto out;
2167 
2168 	err = ext4_generic_delete_entry(handle, dir, de_del,
2169 					bh, bh->b_data,
2170 					dir->i_sb->s_blocksize, csum_size);
2171 	if (err)
2172 		goto out;
2173 
2174 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2175 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2176 	if (unlikely(err))
2177 		goto out;
2178 
2179 	return 0;
2180 out:
2181 	if (err != -ENOENT)
2182 		ext4_std_error(dir->i_sb, err);
2183 	return err;
2184 }
2185 
2186 /*
2187  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2188  * since this indicates that nlinks count was previously 1.
2189  */
2190 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2191 {
2192 	inc_nlink(inode);
2193 	if (is_dx(inode) && inode->i_nlink > 1) {
2194 		/* limit is 16-bit i_links_count */
2195 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2196 			set_nlink(inode, 1);
2197 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2198 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2199 		}
2200 	}
2201 }
2202 
2203 /*
2204  * If a directory had nlink == 1, then we should let it be 1. This indicates
2205  * directory has >EXT4_LINK_MAX subdirs.
2206  */
2207 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2208 {
2209 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2210 		drop_nlink(inode);
2211 }
2212 
2213 
2214 static int ext4_add_nondir(handle_t *handle,
2215 		struct dentry *dentry, struct inode *inode)
2216 {
2217 	int err = ext4_add_entry(handle, dentry, inode);
2218 	if (!err) {
2219 		ext4_mark_inode_dirty(handle, inode);
2220 		unlock_new_inode(inode);
2221 		d_instantiate(dentry, inode);
2222 		return 0;
2223 	}
2224 	drop_nlink(inode);
2225 	unlock_new_inode(inode);
2226 	iput(inode);
2227 	return err;
2228 }
2229 
2230 /*
2231  * By the time this is called, we already have created
2232  * the directory cache entry for the new file, but it
2233  * is so far negative - it has no inode.
2234  *
2235  * If the create succeeds, we fill in the inode information
2236  * with d_instantiate().
2237  */
2238 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2239 		       bool excl)
2240 {
2241 	handle_t *handle;
2242 	struct inode *inode;
2243 	int err, credits, retries = 0;
2244 
2245 	dquot_initialize(dir);
2246 
2247 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2248 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2249 retry:
2250 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2251 					    NULL, EXT4_HT_DIR, credits);
2252 	handle = ext4_journal_current_handle();
2253 	err = PTR_ERR(inode);
2254 	if (!IS_ERR(inode)) {
2255 		inode->i_op = &ext4_file_inode_operations;
2256 		inode->i_fop = &ext4_file_operations;
2257 		ext4_set_aops(inode);
2258 		err = ext4_add_nondir(handle, dentry, inode);
2259 		if (!err && IS_DIRSYNC(dir))
2260 			ext4_handle_sync(handle);
2261 	}
2262 	if (handle)
2263 		ext4_journal_stop(handle);
2264 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2265 		goto retry;
2266 	return err;
2267 }
2268 
2269 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2270 		      umode_t mode, dev_t rdev)
2271 {
2272 	handle_t *handle;
2273 	struct inode *inode;
2274 	int err, credits, retries = 0;
2275 
2276 	if (!new_valid_dev(rdev))
2277 		return -EINVAL;
2278 
2279 	dquot_initialize(dir);
2280 
2281 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2282 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2283 retry:
2284 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2285 					    NULL, EXT4_HT_DIR, credits);
2286 	handle = ext4_journal_current_handle();
2287 	err = PTR_ERR(inode);
2288 	if (!IS_ERR(inode)) {
2289 		init_special_inode(inode, inode->i_mode, rdev);
2290 		inode->i_op = &ext4_special_inode_operations;
2291 		err = ext4_add_nondir(handle, dentry, inode);
2292 		if (!err && IS_DIRSYNC(dir))
2293 			ext4_handle_sync(handle);
2294 	}
2295 	if (handle)
2296 		ext4_journal_stop(handle);
2297 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2298 		goto retry;
2299 	return err;
2300 }
2301 
2302 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2303 			  struct ext4_dir_entry_2 *de,
2304 			  int blocksize, int csum_size,
2305 			  unsigned int parent_ino, int dotdot_real_len)
2306 {
2307 	de->inode = cpu_to_le32(inode->i_ino);
2308 	de->name_len = 1;
2309 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2310 					   blocksize);
2311 	strcpy(de->name, ".");
2312 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2313 
2314 	de = ext4_next_entry(de, blocksize);
2315 	de->inode = cpu_to_le32(parent_ino);
2316 	de->name_len = 2;
2317 	if (!dotdot_real_len)
2318 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2319 					(csum_size + EXT4_DIR_REC_LEN(1)),
2320 					blocksize);
2321 	else
2322 		de->rec_len = ext4_rec_len_to_disk(
2323 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2324 	strcpy(de->name, "..");
2325 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2326 
2327 	return ext4_next_entry(de, blocksize);
2328 }
2329 
2330 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2331 			     struct inode *inode)
2332 {
2333 	struct buffer_head *dir_block = NULL;
2334 	struct ext4_dir_entry_2 *de;
2335 	struct ext4_dir_entry_tail *t;
2336 	ext4_lblk_t block = 0;
2337 	unsigned int blocksize = dir->i_sb->s_blocksize;
2338 	int csum_size = 0;
2339 	int err;
2340 
2341 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2342 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2343 		csum_size = sizeof(struct ext4_dir_entry_tail);
2344 
2345 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2346 		err = ext4_try_create_inline_dir(handle, dir, inode);
2347 		if (err < 0 && err != -ENOSPC)
2348 			goto out;
2349 		if (!err)
2350 			goto out;
2351 	}
2352 
2353 	inode->i_size = 0;
2354 	dir_block = ext4_append(handle, inode, &block);
2355 	if (IS_ERR(dir_block))
2356 		return PTR_ERR(dir_block);
2357 	BUFFER_TRACE(dir_block, "get_write_access");
2358 	err = ext4_journal_get_write_access(handle, dir_block);
2359 	if (err)
2360 		goto out;
2361 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2362 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2363 	set_nlink(inode, 2);
2364 	if (csum_size) {
2365 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2366 		initialize_dirent_tail(t, blocksize);
2367 	}
2368 
2369 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2370 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2371 	if (err)
2372 		goto out;
2373 	set_buffer_verified(dir_block);
2374 out:
2375 	brelse(dir_block);
2376 	return err;
2377 }
2378 
2379 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2380 {
2381 	handle_t *handle;
2382 	struct inode *inode;
2383 	int err, credits, retries = 0;
2384 
2385 	if (EXT4_DIR_LINK_MAX(dir))
2386 		return -EMLINK;
2387 
2388 	dquot_initialize(dir);
2389 
2390 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2391 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2392 retry:
2393 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2394 					    &dentry->d_name,
2395 					    0, NULL, EXT4_HT_DIR, credits);
2396 	handle = ext4_journal_current_handle();
2397 	err = PTR_ERR(inode);
2398 	if (IS_ERR(inode))
2399 		goto out_stop;
2400 
2401 	inode->i_op = &ext4_dir_inode_operations;
2402 	inode->i_fop = &ext4_dir_operations;
2403 	err = ext4_init_new_dir(handle, dir, inode);
2404 	if (err)
2405 		goto out_clear_inode;
2406 	err = ext4_mark_inode_dirty(handle, inode);
2407 	if (!err)
2408 		err = ext4_add_entry(handle, dentry, inode);
2409 	if (err) {
2410 out_clear_inode:
2411 		clear_nlink(inode);
2412 		unlock_new_inode(inode);
2413 		ext4_mark_inode_dirty(handle, inode);
2414 		iput(inode);
2415 		goto out_stop;
2416 	}
2417 	ext4_inc_count(handle, dir);
2418 	ext4_update_dx_flag(dir);
2419 	err = ext4_mark_inode_dirty(handle, dir);
2420 	if (err)
2421 		goto out_clear_inode;
2422 	unlock_new_inode(inode);
2423 	d_instantiate(dentry, inode);
2424 	if (IS_DIRSYNC(dir))
2425 		ext4_handle_sync(handle);
2426 
2427 out_stop:
2428 	if (handle)
2429 		ext4_journal_stop(handle);
2430 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2431 		goto retry;
2432 	return err;
2433 }
2434 
2435 /*
2436  * routine to check that the specified directory is empty (for rmdir)
2437  */
2438 static int empty_dir(struct inode *inode)
2439 {
2440 	unsigned int offset;
2441 	struct buffer_head *bh;
2442 	struct ext4_dir_entry_2 *de, *de1;
2443 	struct super_block *sb;
2444 	int err = 0;
2445 
2446 	if (ext4_has_inline_data(inode)) {
2447 		int has_inline_data = 1;
2448 
2449 		err = empty_inline_dir(inode, &has_inline_data);
2450 		if (has_inline_data)
2451 			return err;
2452 	}
2453 
2454 	sb = inode->i_sb;
2455 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2456 		EXT4_ERROR_INODE(inode, "invalid size");
2457 		return 1;
2458 	}
2459 	bh = ext4_read_dirblock(inode, 0, EITHER);
2460 	if (IS_ERR(bh))
2461 		return 1;
2462 
2463 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2464 	de1 = ext4_next_entry(de, sb->s_blocksize);
2465 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2466 			!le32_to_cpu(de1->inode) ||
2467 			strcmp(".", de->name) ||
2468 			strcmp("..", de1->name)) {
2469 		ext4_warning(inode->i_sb,
2470 			     "bad directory (dir #%lu) - no `.' or `..'",
2471 			     inode->i_ino);
2472 		brelse(bh);
2473 		return 1;
2474 	}
2475 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2476 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2477 	de = ext4_next_entry(de1, sb->s_blocksize);
2478 	while (offset < inode->i_size) {
2479 		if (!bh ||
2480 		    (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2481 			unsigned int lblock;
2482 			err = 0;
2483 			brelse(bh);
2484 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2485 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2486 			if (IS_ERR(bh))
2487 				return 1;
2488 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2489 		}
2490 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2491 					 bh->b_data, bh->b_size, offset)) {
2492 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2493 							 sb->s_blocksize);
2494 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2495 			continue;
2496 		}
2497 		if (le32_to_cpu(de->inode)) {
2498 			brelse(bh);
2499 			return 0;
2500 		}
2501 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2502 		de = ext4_next_entry(de, sb->s_blocksize);
2503 	}
2504 	brelse(bh);
2505 	return 1;
2506 }
2507 
2508 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2509  * such inodes, starting at the superblock, in case we crash before the
2510  * file is closed/deleted, or in case the inode truncate spans multiple
2511  * transactions and the last transaction is not recovered after a crash.
2512  *
2513  * At filesystem recovery time, we walk this list deleting unlinked
2514  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2515  */
2516 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2517 {
2518 	struct super_block *sb = inode->i_sb;
2519 	struct ext4_iloc iloc;
2520 	int err = 0, rc;
2521 
2522 	if (!EXT4_SB(sb)->s_journal)
2523 		return 0;
2524 
2525 	mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2526 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2527 		goto out_unlock;
2528 
2529 	/*
2530 	 * Orphan handling is only valid for files with data blocks
2531 	 * being truncated, or files being unlinked. Note that we either
2532 	 * hold i_mutex, or the inode can not be referenced from outside,
2533 	 * so i_nlink should not be bumped due to race
2534 	 */
2535 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2536 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2537 
2538 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2539 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2540 	if (err)
2541 		goto out_unlock;
2542 
2543 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2544 	if (err)
2545 		goto out_unlock;
2546 	/*
2547 	 * Due to previous errors inode may be already a part of on-disk
2548 	 * orphan list. If so skip on-disk list modification.
2549 	 */
2550 	if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2551 		(le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2552 			goto mem_insert;
2553 
2554 	/* Insert this inode at the head of the on-disk orphan list... */
2555 	NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2556 	EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2557 	err = ext4_handle_dirty_super(handle, sb);
2558 	rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2559 	if (!err)
2560 		err = rc;
2561 
2562 	/* Only add to the head of the in-memory list if all the
2563 	 * previous operations succeeded.  If the orphan_add is going to
2564 	 * fail (possibly taking the journal offline), we can't risk
2565 	 * leaving the inode on the orphan list: stray orphan-list
2566 	 * entries can cause panics at unmount time.
2567 	 *
2568 	 * This is safe: on error we're going to ignore the orphan list
2569 	 * anyway on the next recovery. */
2570 mem_insert:
2571 	if (!err)
2572 		list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2573 
2574 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2575 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2576 			inode->i_ino, NEXT_ORPHAN(inode));
2577 out_unlock:
2578 	mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2579 	ext4_std_error(inode->i_sb, err);
2580 	return err;
2581 }
2582 
2583 /*
2584  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2585  * of such inodes stored on disk, because it is finally being cleaned up.
2586  */
2587 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2588 {
2589 	struct list_head *prev;
2590 	struct ext4_inode_info *ei = EXT4_I(inode);
2591 	struct ext4_sb_info *sbi;
2592 	__u32 ino_next;
2593 	struct ext4_iloc iloc;
2594 	int err = 0;
2595 
2596 	if ((!EXT4_SB(inode->i_sb)->s_journal) &&
2597 	    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2598 		return 0;
2599 
2600 	mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2601 	if (list_empty(&ei->i_orphan))
2602 		goto out;
2603 
2604 	ino_next = NEXT_ORPHAN(inode);
2605 	prev = ei->i_orphan.prev;
2606 	sbi = EXT4_SB(inode->i_sb);
2607 
2608 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2609 
2610 	list_del_init(&ei->i_orphan);
2611 
2612 	/* If we're on an error path, we may not have a valid
2613 	 * transaction handle with which to update the orphan list on
2614 	 * disk, but we still need to remove the inode from the linked
2615 	 * list in memory. */
2616 	if (!handle)
2617 		goto out;
2618 
2619 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2620 	if (err)
2621 		goto out_err;
2622 
2623 	if (prev == &sbi->s_orphan) {
2624 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2625 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2626 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2627 		if (err)
2628 			goto out_brelse;
2629 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2630 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2631 	} else {
2632 		struct ext4_iloc iloc2;
2633 		struct inode *i_prev =
2634 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2635 
2636 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2637 			  i_prev->i_ino, ino_next);
2638 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2639 		if (err)
2640 			goto out_brelse;
2641 		NEXT_ORPHAN(i_prev) = ino_next;
2642 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2643 	}
2644 	if (err)
2645 		goto out_brelse;
2646 	NEXT_ORPHAN(inode) = 0;
2647 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2648 
2649 out_err:
2650 	ext4_std_error(inode->i_sb, err);
2651 out:
2652 	mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2653 	return err;
2654 
2655 out_brelse:
2656 	brelse(iloc.bh);
2657 	goto out_err;
2658 }
2659 
2660 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2661 {
2662 	int retval;
2663 	struct inode *inode;
2664 	struct buffer_head *bh;
2665 	struct ext4_dir_entry_2 *de;
2666 	handle_t *handle = NULL;
2667 
2668 	/* Initialize quotas before so that eventual writes go in
2669 	 * separate transaction */
2670 	dquot_initialize(dir);
2671 	dquot_initialize(dentry->d_inode);
2672 
2673 	retval = -ENOENT;
2674 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2675 	if (!bh)
2676 		goto end_rmdir;
2677 
2678 	inode = dentry->d_inode;
2679 
2680 	retval = -EIO;
2681 	if (le32_to_cpu(de->inode) != inode->i_ino)
2682 		goto end_rmdir;
2683 
2684 	retval = -ENOTEMPTY;
2685 	if (!empty_dir(inode))
2686 		goto end_rmdir;
2687 
2688 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2689 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2690 	if (IS_ERR(handle)) {
2691 		retval = PTR_ERR(handle);
2692 		handle = NULL;
2693 		goto end_rmdir;
2694 	}
2695 
2696 	if (IS_DIRSYNC(dir))
2697 		ext4_handle_sync(handle);
2698 
2699 	retval = ext4_delete_entry(handle, dir, de, bh);
2700 	if (retval)
2701 		goto end_rmdir;
2702 	if (!EXT4_DIR_LINK_EMPTY(inode))
2703 		ext4_warning(inode->i_sb,
2704 			     "empty directory has too many links (%d)",
2705 			     inode->i_nlink);
2706 	inode->i_version++;
2707 	clear_nlink(inode);
2708 	/* There's no need to set i_disksize: the fact that i_nlink is
2709 	 * zero will ensure that the right thing happens during any
2710 	 * recovery. */
2711 	inode->i_size = 0;
2712 	ext4_orphan_add(handle, inode);
2713 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2714 	ext4_mark_inode_dirty(handle, inode);
2715 	ext4_dec_count(handle, dir);
2716 	ext4_update_dx_flag(dir);
2717 	ext4_mark_inode_dirty(handle, dir);
2718 
2719 end_rmdir:
2720 	brelse(bh);
2721 	if (handle)
2722 		ext4_journal_stop(handle);
2723 	return retval;
2724 }
2725 
2726 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2727 {
2728 	int retval;
2729 	struct inode *inode;
2730 	struct buffer_head *bh;
2731 	struct ext4_dir_entry_2 *de;
2732 	handle_t *handle = NULL;
2733 
2734 	trace_ext4_unlink_enter(dir, dentry);
2735 	/* Initialize quotas before so that eventual writes go
2736 	 * in separate transaction */
2737 	dquot_initialize(dir);
2738 	dquot_initialize(dentry->d_inode);
2739 
2740 	retval = -ENOENT;
2741 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2742 	if (!bh)
2743 		goto end_unlink;
2744 
2745 	inode = dentry->d_inode;
2746 
2747 	retval = -EIO;
2748 	if (le32_to_cpu(de->inode) != inode->i_ino)
2749 		goto end_unlink;
2750 
2751 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2752 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2753 	if (IS_ERR(handle)) {
2754 		retval = PTR_ERR(handle);
2755 		handle = NULL;
2756 		goto end_unlink;
2757 	}
2758 
2759 	if (IS_DIRSYNC(dir))
2760 		ext4_handle_sync(handle);
2761 
2762 	if (!inode->i_nlink) {
2763 		ext4_warning(inode->i_sb,
2764 			     "Deleting nonexistent file (%lu), %d",
2765 			     inode->i_ino, inode->i_nlink);
2766 		set_nlink(inode, 1);
2767 	}
2768 	retval = ext4_delete_entry(handle, dir, de, bh);
2769 	if (retval)
2770 		goto end_unlink;
2771 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2772 	ext4_update_dx_flag(dir);
2773 	ext4_mark_inode_dirty(handle, dir);
2774 	drop_nlink(inode);
2775 	if (!inode->i_nlink)
2776 		ext4_orphan_add(handle, inode);
2777 	inode->i_ctime = ext4_current_time(inode);
2778 	ext4_mark_inode_dirty(handle, inode);
2779 	retval = 0;
2780 
2781 end_unlink:
2782 	brelse(bh);
2783 	if (handle)
2784 		ext4_journal_stop(handle);
2785 	trace_ext4_unlink_exit(dentry, retval);
2786 	return retval;
2787 }
2788 
2789 static int ext4_symlink(struct inode *dir,
2790 			struct dentry *dentry, const char *symname)
2791 {
2792 	handle_t *handle;
2793 	struct inode *inode;
2794 	int l, err, retries = 0;
2795 	int credits;
2796 
2797 	l = strlen(symname)+1;
2798 	if (l > dir->i_sb->s_blocksize)
2799 		return -ENAMETOOLONG;
2800 
2801 	dquot_initialize(dir);
2802 
2803 	if (l > EXT4_N_BLOCKS * 4) {
2804 		/*
2805 		 * For non-fast symlinks, we just allocate inode and put it on
2806 		 * orphan list in the first transaction => we need bitmap,
2807 		 * group descriptor, sb, inode block, quota blocks, and
2808 		 * possibly selinux xattr blocks.
2809 		 */
2810 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2811 			  EXT4_XATTR_TRANS_BLOCKS;
2812 	} else {
2813 		/*
2814 		 * Fast symlink. We have to add entry to directory
2815 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2816 		 * allocate new inode (bitmap, group descriptor, inode block,
2817 		 * quota blocks, sb is already counted in previous macros).
2818 		 */
2819 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2820 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2821 	}
2822 retry:
2823 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2824 					    &dentry->d_name, 0, NULL,
2825 					    EXT4_HT_DIR, credits);
2826 	handle = ext4_journal_current_handle();
2827 	err = PTR_ERR(inode);
2828 	if (IS_ERR(inode))
2829 		goto out_stop;
2830 
2831 	if (l > EXT4_N_BLOCKS * 4) {
2832 		inode->i_op = &ext4_symlink_inode_operations;
2833 		ext4_set_aops(inode);
2834 		/*
2835 		 * We cannot call page_symlink() with transaction started
2836 		 * because it calls into ext4_write_begin() which can wait
2837 		 * for transaction commit if we are running out of space
2838 		 * and thus we deadlock. So we have to stop transaction now
2839 		 * and restart it when symlink contents is written.
2840 		 *
2841 		 * To keep fs consistent in case of crash, we have to put inode
2842 		 * to orphan list in the mean time.
2843 		 */
2844 		drop_nlink(inode);
2845 		err = ext4_orphan_add(handle, inode);
2846 		ext4_journal_stop(handle);
2847 		if (err)
2848 			goto err_drop_inode;
2849 		err = __page_symlink(inode, symname, l, 1);
2850 		if (err)
2851 			goto err_drop_inode;
2852 		/*
2853 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2854 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2855 		 */
2856 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
2857 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2858 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2859 		if (IS_ERR(handle)) {
2860 			err = PTR_ERR(handle);
2861 			goto err_drop_inode;
2862 		}
2863 		set_nlink(inode, 1);
2864 		err = ext4_orphan_del(handle, inode);
2865 		if (err) {
2866 			ext4_journal_stop(handle);
2867 			clear_nlink(inode);
2868 			goto err_drop_inode;
2869 		}
2870 	} else {
2871 		/* clear the extent format for fast symlink */
2872 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2873 		inode->i_op = &ext4_fast_symlink_inode_operations;
2874 		memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2875 		inode->i_size = l-1;
2876 	}
2877 	EXT4_I(inode)->i_disksize = inode->i_size;
2878 	err = ext4_add_nondir(handle, dentry, inode);
2879 	if (!err && IS_DIRSYNC(dir))
2880 		ext4_handle_sync(handle);
2881 
2882 out_stop:
2883 	if (handle)
2884 		ext4_journal_stop(handle);
2885 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2886 		goto retry;
2887 	return err;
2888 err_drop_inode:
2889 	unlock_new_inode(inode);
2890 	iput(inode);
2891 	return err;
2892 }
2893 
2894 static int ext4_link(struct dentry *old_dentry,
2895 		     struct inode *dir, struct dentry *dentry)
2896 {
2897 	handle_t *handle;
2898 	struct inode *inode = old_dentry->d_inode;
2899 	int err, retries = 0;
2900 
2901 	if (inode->i_nlink >= EXT4_LINK_MAX)
2902 		return -EMLINK;
2903 
2904 	dquot_initialize(dir);
2905 
2906 retry:
2907 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2908 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2909 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS));
2910 	if (IS_ERR(handle))
2911 		return PTR_ERR(handle);
2912 
2913 	if (IS_DIRSYNC(dir))
2914 		ext4_handle_sync(handle);
2915 
2916 	inode->i_ctime = ext4_current_time(inode);
2917 	ext4_inc_count(handle, inode);
2918 	ihold(inode);
2919 
2920 	err = ext4_add_entry(handle, dentry, inode);
2921 	if (!err) {
2922 		ext4_mark_inode_dirty(handle, inode);
2923 		d_instantiate(dentry, inode);
2924 	} else {
2925 		drop_nlink(inode);
2926 		iput(inode);
2927 	}
2928 	ext4_journal_stop(handle);
2929 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2930 		goto retry;
2931 	return err;
2932 }
2933 
2934 
2935 /*
2936  * Try to find buffer head where contains the parent block.
2937  * It should be the inode block if it is inlined or the 1st block
2938  * if it is a normal dir.
2939  */
2940 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2941 					struct inode *inode,
2942 					int *retval,
2943 					struct ext4_dir_entry_2 **parent_de,
2944 					int *inlined)
2945 {
2946 	struct buffer_head *bh;
2947 
2948 	if (!ext4_has_inline_data(inode)) {
2949 		bh = ext4_read_dirblock(inode, 0, EITHER);
2950 		if (IS_ERR(bh)) {
2951 			*retval = PTR_ERR(bh);
2952 			return NULL;
2953 		}
2954 		*parent_de = ext4_next_entry(
2955 					(struct ext4_dir_entry_2 *)bh->b_data,
2956 					inode->i_sb->s_blocksize);
2957 		return bh;
2958 	}
2959 
2960 	*inlined = 1;
2961 	return ext4_get_first_inline_block(inode, parent_de, retval);
2962 }
2963 
2964 /*
2965  * Anybody can rename anything with this: the permission checks are left to the
2966  * higher-level routines.
2967  */
2968 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2969 		       struct inode *new_dir, struct dentry *new_dentry)
2970 {
2971 	handle_t *handle;
2972 	struct inode *old_inode, *new_inode;
2973 	struct buffer_head *old_bh, *new_bh, *dir_bh;
2974 	struct ext4_dir_entry_2 *old_de, *new_de;
2975 	int retval, force_da_alloc = 0;
2976 	int inlined = 0, new_inlined = 0;
2977 	struct ext4_dir_entry_2 *parent_de;
2978 
2979 	dquot_initialize(old_dir);
2980 	dquot_initialize(new_dir);
2981 
2982 	old_bh = new_bh = dir_bh = NULL;
2983 
2984 	/* Initialize quotas before so that eventual writes go
2985 	 * in separate transaction */
2986 	if (new_dentry->d_inode)
2987 		dquot_initialize(new_dentry->d_inode);
2988 	handle = ext4_journal_start(old_dir, EXT4_HT_DIR,
2989 		(2 * EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2990 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
2991 	if (IS_ERR(handle))
2992 		return PTR_ERR(handle);
2993 
2994 	if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2995 		ext4_handle_sync(handle);
2996 
2997 	old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
2998 	/*
2999 	 *  Check for inode number is _not_ due to possible IO errors.
3000 	 *  We might rmdir the source, keep it as pwd of some process
3001 	 *  and merrily kill the link to whatever was created under the
3002 	 *  same name. Goodbye sticky bit ;-<
3003 	 */
3004 	old_inode = old_dentry->d_inode;
3005 	retval = -ENOENT;
3006 	if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
3007 		goto end_rename;
3008 
3009 	new_inode = new_dentry->d_inode;
3010 	new_bh = ext4_find_entry(new_dir, &new_dentry->d_name,
3011 				 &new_de, &new_inlined);
3012 	if (new_bh) {
3013 		if (!new_inode) {
3014 			brelse(new_bh);
3015 			new_bh = NULL;
3016 		}
3017 	}
3018 	if (S_ISDIR(old_inode->i_mode)) {
3019 		if (new_inode) {
3020 			retval = -ENOTEMPTY;
3021 			if (!empty_dir(new_inode))
3022 				goto end_rename;
3023 		}
3024 		retval = -EIO;
3025 		dir_bh = ext4_get_first_dir_block(handle, old_inode,
3026 						  &retval, &parent_de,
3027 						  &inlined);
3028 		if (!dir_bh)
3029 			goto end_rename;
3030 		if (le32_to_cpu(parent_de->inode) != old_dir->i_ino)
3031 			goto end_rename;
3032 		retval = -EMLINK;
3033 		if (!new_inode && new_dir != old_dir &&
3034 		    EXT4_DIR_LINK_MAX(new_dir))
3035 			goto end_rename;
3036 		BUFFER_TRACE(dir_bh, "get_write_access");
3037 		retval = ext4_journal_get_write_access(handle, dir_bh);
3038 		if (retval)
3039 			goto end_rename;
3040 	}
3041 	if (!new_bh) {
3042 		retval = ext4_add_entry(handle, new_dentry, old_inode);
3043 		if (retval)
3044 			goto end_rename;
3045 	} else {
3046 		BUFFER_TRACE(new_bh, "get write access");
3047 		retval = ext4_journal_get_write_access(handle, new_bh);
3048 		if (retval)
3049 			goto end_rename;
3050 		new_de->inode = cpu_to_le32(old_inode->i_ino);
3051 		if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3052 					      EXT4_FEATURE_INCOMPAT_FILETYPE))
3053 			new_de->file_type = old_de->file_type;
3054 		new_dir->i_version++;
3055 		new_dir->i_ctime = new_dir->i_mtime =
3056 					ext4_current_time(new_dir);
3057 		ext4_mark_inode_dirty(handle, new_dir);
3058 		BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3059 		if (!new_inlined) {
3060 			retval = ext4_handle_dirty_dirent_node(handle,
3061 							       new_dir, new_bh);
3062 			if (unlikely(retval)) {
3063 				ext4_std_error(new_dir->i_sb, retval);
3064 				goto end_rename;
3065 			}
3066 		}
3067 		brelse(new_bh);
3068 		new_bh = NULL;
3069 	}
3070 
3071 	/*
3072 	 * Like most other Unix systems, set the ctime for inodes on a
3073 	 * rename.
3074 	 */
3075 	old_inode->i_ctime = ext4_current_time(old_inode);
3076 	ext4_mark_inode_dirty(handle, old_inode);
3077 
3078 	/*
3079 	 * ok, that's it
3080 	 */
3081 	if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3082 	    old_de->name_len != old_dentry->d_name.len ||
3083 	    strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3084 	    (retval = ext4_delete_entry(handle, old_dir,
3085 					old_de, old_bh)) == -ENOENT) {
3086 		/* old_de could have moved from under us during htree split, so
3087 		 * make sure that we are deleting the right entry.  We might
3088 		 * also be pointing to a stale entry in the unused part of
3089 		 * old_bh so just checking inum and the name isn't enough. */
3090 		struct buffer_head *old_bh2;
3091 		struct ext4_dir_entry_2 *old_de2;
3092 
3093 		old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
3094 					  &old_de2, NULL);
3095 		if (old_bh2) {
3096 			retval = ext4_delete_entry(handle, old_dir,
3097 						   old_de2, old_bh2);
3098 			brelse(old_bh2);
3099 		}
3100 	}
3101 	if (retval) {
3102 		ext4_warning(old_dir->i_sb,
3103 				"Deleting old file (%lu), %d, error=%d",
3104 				old_dir->i_ino, old_dir->i_nlink, retval);
3105 	}
3106 
3107 	if (new_inode) {
3108 		ext4_dec_count(handle, new_inode);
3109 		new_inode->i_ctime = ext4_current_time(new_inode);
3110 	}
3111 	old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3112 	ext4_update_dx_flag(old_dir);
3113 	if (dir_bh) {
3114 		parent_de->inode = cpu_to_le32(new_dir->i_ino);
3115 		BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3116 		if (!inlined) {
3117 			if (is_dx(old_inode)) {
3118 				retval = ext4_handle_dirty_dx_node(handle,
3119 								   old_inode,
3120 								   dir_bh);
3121 			} else {
3122 				retval = ext4_handle_dirty_dirent_node(handle,
3123 							old_inode, dir_bh);
3124 			}
3125 		} else {
3126 			retval = ext4_mark_inode_dirty(handle, old_inode);
3127 		}
3128 		if (retval) {
3129 			ext4_std_error(old_dir->i_sb, retval);
3130 			goto end_rename;
3131 		}
3132 		ext4_dec_count(handle, old_dir);
3133 		if (new_inode) {
3134 			/* checked empty_dir above, can't have another parent,
3135 			 * ext4_dec_count() won't work for many-linked dirs */
3136 			clear_nlink(new_inode);
3137 		} else {
3138 			ext4_inc_count(handle, new_dir);
3139 			ext4_update_dx_flag(new_dir);
3140 			ext4_mark_inode_dirty(handle, new_dir);
3141 		}
3142 	}
3143 	ext4_mark_inode_dirty(handle, old_dir);
3144 	if (new_inode) {
3145 		ext4_mark_inode_dirty(handle, new_inode);
3146 		if (!new_inode->i_nlink)
3147 			ext4_orphan_add(handle, new_inode);
3148 		if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3149 			force_da_alloc = 1;
3150 	}
3151 	retval = 0;
3152 
3153 end_rename:
3154 	brelse(dir_bh);
3155 	brelse(old_bh);
3156 	brelse(new_bh);
3157 	ext4_journal_stop(handle);
3158 	if (retval == 0 && force_da_alloc)
3159 		ext4_alloc_da_blocks(old_inode);
3160 	return retval;
3161 }
3162 
3163 /*
3164  * directories can handle most operations...
3165  */
3166 const struct inode_operations ext4_dir_inode_operations = {
3167 	.create		= ext4_create,
3168 	.lookup		= ext4_lookup,
3169 	.link		= ext4_link,
3170 	.unlink		= ext4_unlink,
3171 	.symlink	= ext4_symlink,
3172 	.mkdir		= ext4_mkdir,
3173 	.rmdir		= ext4_rmdir,
3174 	.mknod		= ext4_mknod,
3175 	.rename		= ext4_rename,
3176 	.setattr	= ext4_setattr,
3177 	.setxattr	= generic_setxattr,
3178 	.getxattr	= generic_getxattr,
3179 	.listxattr	= ext4_listxattr,
3180 	.removexattr	= generic_removexattr,
3181 	.get_acl	= ext4_get_acl,
3182 	.fiemap         = ext4_fiemap,
3183 };
3184 
3185 const struct inode_operations ext4_special_inode_operations = {
3186 	.setattr	= ext4_setattr,
3187 	.setxattr	= generic_setxattr,
3188 	.getxattr	= generic_getxattr,
3189 	.listxattr	= ext4_listxattr,
3190 	.removexattr	= generic_removexattr,
3191 	.get_acl	= ext4_get_acl,
3192 };
3193